GDG AI for Science - Australia
Let's explore the AlphaGenome pre-print together and learn how this AI tool can decipher the regulatory code within DNA sequences and predict how single genetic variations impact biological processes.
87 RSVP'd
Join the GDG AI for Science community to discuss AlphaGenome, a deep learning model from Google DeepMind that predicts how different parts of DNA are controlled and how they function to regulate genes. AlphaGenome analyses long DNA sequences (up to 1 million base-pairs) to predict thousands of diverse molecular properties characterising their regulatory activity at single base-pair resolution, fundamentally advancing regulatory variant-effect prediction. This helps us understand the impact of tiny changes (variants) in our DNA, specifically in the regions that control when and how our genes are turned on or off.
In this collaborative session, you will help unpack the paper to understand:
What AlphaGenome is: A model that predicts diverse genomic data types, including gene expression, chromatin accessibility, TF binding, splicing patterns, and 3D chromatin architecture.
How it works: We can touch on the U-Net-inspired architecture, which uses transformers to capture long-range interactions, and its two-stage pre-training and distillation process.
What AlphaGenome is capable of: AlphaGenome matches or exceeds the performance of specialised models on most benchmarks and can accurately recapitulate the complex mechanisms of clinically-relevant variants.
How to use it: We will run through a hands-on Python exercise to demonstrate how to use AlphaGenome to score the impact of a genetic variant in real-time.
What it CANNOT do: Let's dig into any issues with the paper and understand the limitations of AlphaGenome.
Whatever else?
This is a virtual facilitated session, interaction and active participation is encouraged. Discussion is aimed to go where we deem interesting.
This event is for anyone interested in the intersection of artificial intelligence and the life sciences, including researchers, developers, students, bioinformaticians, computational biologists, scientific support staff, and clinicians.
Prerequisites:
Read the DeepMind blog post.
Read the pre-print paper.
Have a Google account to follow along examples in Google Colab.
Get an AlphaGenome API key (we can cover this).
Fill in our community survey to express your ideas.
Science Catalyst Program Manager
The University of Queensland
Senior Research Fellow
Monash University
Organizer
Haizea Analytics
Organizer
Monash University
Organizer
Macquarie University
Macquarie University
The University of Queensland
University of Queensland
Science Catalyst Program Manager
University of Sydney
University of Sydney
Contact Us